BioSystems 100 (2010) 1–7 Contents lists available at ScienceDirect BioSystems journal homepage: www.elsevier.com/locate/biosystems In vitro molecular pattern classification via DNA-based weighted-sum operation Hee-Woong Lim a,1 , Seung Hwan Lee c,1 , Kyung-Ae Yang b , Ji Youn Lee d , Suk-In Yoo b , Tai Hyun Park c , Byoung-Tak Zhang b,∗ a Department of Computer Science and Engineering, Ewha Womans University, Seoul 120-750, Republic of Korea b School of Computer Science and Engineering and the Center for Biointelligence Technology (CBIT), Seoul National University, Seoul 151-744, Republic of Korea c School of Chemical and Biological Engineering, Seoul National University, Seoul 151-744, Republic of Korea d Department of Biomedical Engineering, University of California, Davis, Davis, CA 95616, USA article info Article history: Received 4 June 2009 Received in revised form 30 November 2009 Accepted 2 December 2009 Keywords: Molecular pattern classification DNA computing DNA-based weighted-sum operation Competitive hybridization abstract Recent progress in molecular computation suggests the possibility of pattern classification in vitro. Weighted sum is a primitive operation required by many pattern classification problems. Here we present a DNA-based molecular computation method for implementing the weighted-sum operation and its use for molecular pattern classification in a test tube. The weights of the classifier are encoded as the mix- ing ratios of the differentially labeled probe DNA molecules, which are competitively hybridized with the input-encoding target molecules to compute the decision boundary of classification. The computa- tion result is detected by fluorescence signals. We experimentally verify the underlying weight encoding scheme and demonstrate successful discrimination of two-group labels of synthetic DNA mixture pat- terns. The method can be used for direct computation on biomolecular data in a liquid state. © 2009 Elsevier Ireland Ltd. All rights reserved. 1. Introduction The identification of biomarkers and their analysis for disease diagnosis have recently emerged as important issues. Many of those problems involve pattern classification (Bishop, 2006), where the target patterns are represented as mixtures of biomarkers such as messenger RNAs (mRNA), proteins, or microRNAs (miRNA), and their expression levels provide informative features for the classi- fication (Khan et al., 2001; Ramaswamy et al., 2001; Lu et al., 2005). Whereas the conventional analysis methods for these biomolecular patterns require quantitative detection in vitro prior to the analy- sis in silico, recent progress in molecular information processing technology suggests that direct computation on such biochemi- cal information in vitro is realizable (Paun et al., 1998); examples include gene expression analysis and control, extraction of molec- ular features, and algebraic operations in vitro (Oliver, 1997; Mills et al., 1999; Sakakibara and Suyama, 2000; Mills, 2002; Benenson et al., 2004; Lim et al., 2004). Many DNA computing approaches have been developed pre- viously for solving computational problems or for implementing ∗ Corresponding author. E-mail addresses: hwlim@ewha.ac.kr (H.-W. Lim), skulsh78@snu.ac.kr (S.H. Lee), kayang@bi.snu.ac.kr (K.-A. Yang), jiyounlee@gmail.com (J.Y. Lee), siyoo@ailab.snu.ac.kr (S.-I. Yoo), thpark@snu.ac.kr (T.H. Park), btzhang@bi.snu.ac.kr (B.-T. Zhang). 1 These authors contributed equally to this work. Boolean logic circuits in vitro (Henkel et al., 2007; Bakar et al., 2008; Wang et al., 2008; Zoraida et al., 2009). However, more advanced analysis of biochemical information in vitro requires efficient arithmetic operations capable of handling molecular quan- titative information. Although a few theoretically sound models of DNA-based algebraic operations have been proposed (Oliver, 1997; Mills et al., 1999, 2001), they have not yet been imple- mented experimentally. Here, instead, we focus on a simple but significant primitive molecular algebraic operation and propose a molecular pattern classification model that is implemented in DNA computing. Operating in vitro, it directly takes biological inputs such as DNA or RNA molecules and computes the group indices that are displayed optically as fluorescence signals. As a primitive operation, we define a DNA-based weighted-sum computation and formulate simple competitive hybridization reactions between the input molecules and differentially labeled probe mixtures into the weighted sum of inputs. We present the verification results of our weight encoding scheme and demonstrate experimentally the suc- cessful classification of in vitro patterns of synthetic DNA mixtures. This paper is an extended written version of the oral presentation given at the 13th international meeting on DNA computing (Lim et al., 2007). The paper is organized as follows. In Section 2, we describe the basic architecture and theoretical background of the molecular pattern classifier. Section 3 gives a DNA-based implementation of the molecular pattern classifier in vitro including the DNA-based weighted-sum operation. The experimental results are presented in Section 4. Section 5 summarizes the result. More details on mate- 0303-2647/$ – see front matter © 2009 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.biosystems.2009.12.001